0% found this document useful (0 votes)
21 views21 pages

Intelligent Signal Processing and Coordination For The Adaptive

Uploaded by

Petterson Pedro
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
21 views21 pages

Intelligent Signal Processing and Coordination For The Adaptive

Uploaded by

Petterson Pedro
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

©istockphoto.

com/chombosan
Pirathayini Srikantha and Deepa Kundur

Intelligent I
n today’s era of the Internet of Things (IoT),
the amalgamation of information and com-
munication technologies with actuating de-

Signal Processing
vices has reached all corners of the modern
world. In the context of critical infrastructures,
such as the power grid, this cyberphysical trans-
formation has permeated all system levels as evi-

and Coordination
dent in devices ranging from crucial operational
components (e.g., generators) and advanced sen-
sors [e.g., phasor measurement units (PMUs) and
programmable controllers], to consumer-centric

for the Adaptive


devices [smart meters, electric vehicles (EVs),
and smart appliances]. These extended cyber-
physical functionalities have opened up signal
processing opportunities that can be harnessed

Smart Grid
to empower actuating devices to adaptively and
synergistically acquire data, conduct analyt-
ics, and respond to system and environmental
changes for better power-grid operations. In this
article, we demonstrate how a hierarchical signal
An overview of data-driven processing and actuation framework can enable
grid management the tractable all-encompassing coordination of
thousands of actuating power entities to main-
tain efficient operations while accounting for
Digital Object Identifier 10.1109/MSP.2018.2877001
Date of publication: 26 April 2019 physical infrastructure limits.

82 IEEE Signal Processing Magazine | May 2019 | 1053-5888/19©2019IEEE


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
Adaptive monitoring and control years, TNs have come to include solar and wind farms and
Revolutionary changes in clean energy policies along with other IPPs as alternatives to bulk-generation plants based
advances in technologies related to power-consumption and on fossil fuels and associated with large carbon footprints.
power-generation technologies are triggering a major para- Due to the intermittent nature of solar and wind energy,
digm shift in the electricity sector [1]. This shift is blurring TN lines are experiencing unpredictable periods of con-
conventional boundaries in the vertically integrated power gestion that cause the system to function perilously close
grid. With deregulation in electricity mar- to stable limits.
kets and the introduction of sustainabil- The modern grid must Distribution networks (DNs) operate at
ity incentives, independent power plants contend with growing low voltages and tap onto the TN to deliver
(IPPs), consisting of highly fluctuating congestion, inefficiency, power to local consumers. With the signifi-
renewable energy sources, and alternative cant proliferation of distributed generation
and system vulnerabilities
loads, such as EVs, have become more (DG) systems in the DN, sudden surges
common. In increasing numbers, power that can lead to costly in generation can lead to voltage spikes,
consumers are also becoming sellers as equipment damage and, which result in overvoltage conditions.
they deploy microgeneration systems to eventually, cascading Similarly, the deployment of large power-
supplement local electricity needs. Amid failures. consuming equipment, such as EVs in the
this rapidly evolving landscape of electric- DNs, can result in voltage drops that can
ity supply and demand, upgrades to the underlying power lead to undervoltage conditions. These voltage violations can
infrastructure consisting of transmission/distribution lines, trigger passive protection elements, such as fuses and circuit
transformers, and protection devices are falling behind breakers, which can result in outages requiring hours to
[2]. Meanwhile, cyberenabled sensors and actuators are reinstate [4].
­being widely deployed across all levels of the power grid. Hence, the modern grid must contend with growing con-
These resources can be leveraged to enable adaptive moni- gestion, inefficiency, and system vulnerabilities that can lead
toring and control to overcome the arising challenges by cap- to costly equipment damage and, eventually, cascading fail-
italizing on recent advances in data analytics and intelligent ures. Upgrading the grid infrastructure to better accommodate
signal processing techniques. these changes is extremely costly [1]. However, opportunities
lie in the effective use of vast volumes of grid-monitoring data
Ongoing changes in the traditional grid ­generated every second by measuring devices and in offload-
The traditional grid is designed to accommodate a cen- ing decision making to actuating power components by way
tralized power infrastructure in which power flows from of intelligent signal processing. Designing signals that capture
bulk synchronous generation systems to consumers, as il- the general trends in the power grid and efficiently processing
lustrated in Figure 1. The output from these bulk-power these to adaptively compute local actuation decisions enable
plants is predictable, as the prime movers are fueled by more efficient and cost-effective grid operations.
nuclear, gas, coal, or hydroelectric sources directly main-
tained by system operators [3]. High-voltage transmission Opportunities in the cyberphysical power grid
networks (TNs) facilitate the transportation of generated Widespread connectivity permeating modern society has
power across geographically dispersed regions. In recent triggered the IoT era [5]. Devices ranging from wearables

Power Flow

Synchronous Plant

Transformer TN ...

Home

Industry
DN

FIGURE 1. A traditional grid.

IEEE Signal Processing Magazine | May 2019 | 83


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
Storage

...

DG
EV/Storage

Local Communication
Channel Global Communication
Channel

FIGURE 2. The cyberphysical smart grid.

Hierarchical coordination in the smart grid


The cyberphysical power grid is a monolithic entity made
External State
up of millions of active nodes. Coordinating these devices
Actuating Signal
Signal
while accounting for underlying physical grid characteristics
Local Actuating External is an ongoing challenge for researchers due to the immense
Limits Agent Environment number of controllable elements present in the system and
Monitoring the inclusion of nonlinear physical constraints embodying
Data Local State power ­balance and bus/line limits. Moreover, the highly
Signal
varying nature of active nodes, such as the renewables, in-
troduces significant actuation signal uncertainty due to sig-
FIGURE 3. Adaptive actuation. nificant error margins present in the associated long-term
generation-forecast models [6]. Thus, applying the tradition-
al vertically integrated grid-operation model, whereby active
to critical infrastructure elements can communicate, per- elements and signal processing are centrally coordinated, is
form computations locally, and intelligently actuate. As not practical.
illustrated in Figure 2, the modern grid is just one of many The unified hierarchical framework introduced in this
parts of our world where cyberphysical amalgamation—the article presents signal processing mechanisms by which con-
development of interacting networks of computational and centrated computational efforts are offloaded to every actuat-
physical components—is taking place. The vast majority ing element in the system so that they can adaptively react to
of consumer appliances and grid ­management entities, such improve global power-system conditions. For this, key con-
as circuit breakers, relays, transformers, and inverters, are cepts involving decoupling, abstraction, decomposition,
smart. That is, they are capable of decision making facili- and parallel processing are leveraged. Figure 3 illustrates the
tated by signal processing. In addition to these actuating de- fundamental principle used in the design of local decision
vices, many monitoring devices, such as smart meters and making by every active node.
PMUs, are recording the states of the grid in a highly granu- Information about the global state of the system with
lar manner, thus providing opportunities for advanced data respect to highest efficiency and balance in physical con-
analytics. These monitoring and control systems currently in straints is communicated to each cyberenabled node via
place constitute the smart grid [1]. The information-centric signals designed by processing grid-measurement data.
cyberenabled power grid elevates situational awareness and These signals capture summary information regarding the
enables intelligent cyberphysical signal processing-enabled state of the global system. In this article, we distinguish
responses by active nodes to adaptively and effectively react data and signals in the following manner. Raw state mea-
to system changes. surements generated by monitoring devices (e.g., smart

84 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
meters, PMUs) are referred to as data. Signals, mean- present a generalized hierarchical framework with built-in
while, contain insights obtained by strategically combin- flexibility in the way coordination is executed within each
ing and processing these monitoring data. The nodes, upon tier, which represents a particular system being coordi-
receiving these signals, react to improve the external sys- nated. Furthermore, we provide a comprehensive overview
tem state while accounting for local physical constraints. of power-flow equations (including convex relaxations and
These summary signals can be exchanged among active dq0 transformations) in the context of a wide variety of set-
nodes or broadcast to many nodes via dedicated coordinat- tings (such as microgrids and system protection) along with
ing entities. The frequency of signal exchange depends on many practical cases. We also present a broad overview of
the control horizon and tier in which the coordination is coordination paradigms based on centralized, distributed,
taking place. and decentralized techniques that can be flexibly applied to
For tractable analysis, the grid-coordination problem is decom- any tier of the hierarchical framework based on the power
posed into simpler signal processing subproblems based on the entities being coordinated, the system model considered,
control horizon, grid topology, and under- and the control horizon in place.
lying physical network constraints form- The information-centric Hierarchical coordination can be applied
ing tiers within the proposed hierarchical to various components of the power grid.
cyberenabled power For instance, the primary, secondary, and
framework. As illustrated in Figure 4, man-
aging entities representing tiers exchange grid elevates situational tertiary control in place in today’s electric
signals to coordinate the activities of the awareness and enables grid is a hierarchical coordination mecha-
tiers. As these summary signals convey intelligent cyberphysical nism. Primary control deals with real-time
general trends but do not contain specific signal processing-enabled transients and, therefore, is typically decen-
information about individual system enti- responses by active tralized. Secondary coordination involves
ties, abstraction is introduced. This allows the centralized computation of optimal gen-
for the plug-and-play integration of system
nodes to adaptively and eration setpoints. Tertiary control mecha-
components because existing nodes need effectively react to system nisms, such as automatic generation controls
not be aware of the specifics associated changes. (AGCs), are decentralized as local mea-
with these changes to design local actuation surements, and computations are used for
signals. Thus, the hierarchical framework introduces a flexible frequency control over wide areas. Thus, various coordination
mechanism to accommodate the changing landscape of power strategies are employed for the hierarchical management of
demand and supply by leveraging the cyberphysical signal pro- systems ranging from bulk-power grids to microgrids [9].
cessing possible in the modern grid. This hierarchical paradigm can be used to coordinate active
This article differs from our recent work in [7], which nodes at an even more granular manner across a broad range
presents a hierarchical framework for the TN and DN where of control horizons while engaging not only generation enti-
coordination is exacted via two specific methods tailored ties but also active consumers, protective devices, and other
separately for these individual systems. In this article, we actuating entities. Each tier can employ different types of

Planning Tier
Abstraction
Granularity
Increasing

Increasing

Tiers

Physical Actuation Tier

Communication Signals Physical Actuation Coordinated Entity

FIGURE 4. The hierarchical framework.

IEEE Signal Processing Magazine | May 2019 | 85


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
coordination strategies (such as centralized, decentralized, these into the optimal grid coordination problem are high-
or distributed paradigms), which are associated with trad- lighted. Finally, methods that can be leveraged to relax these
eoffs, including control horizon, error margin in forecast relations into tractable forms while retaining defining physi-
models, and presence of a centralized coordinating entity, cal attributes are presented in the context of the TNs, DNs, mi-
to mention a few. The intratier and intertier interactions are crogrids, and power consumers. These tractable formulations
governed by signals designed using underlying system con- can be decomposed into simpler subproblems that can then be
straints of the corresponding tiers and optimization goals. used to glean pertinent information from monitoring data
to design succinct signals reflecting general trends in the
Contributions of this article system. These signals will then be used by actuating nodes to
The effective design of various tiers, signal processing topolo- make adaptive decisions for iteratively achieving reliable and
gies within these tiers, and adaptive decision making entails efficient grid operations.
the following key considerations:
1) structure of the physical infrastructure (i.e., radial DN ver- Power flow and limits
sus mesh TN, low voltage versus high voltage, and so on) In today’s grid, the prevalent mode of operation is ac, in which
2) operational norms and ownership [e.g., deregulation, electric voltages and currents vary in a sinusoidal manner. When no
power utility (EPU), and so on] transience is present, the system operates under what is known
3) signal processing and actuation horizon (e.g., fast response as steady-state conditions, and this is the grid mode used by
for highly fluctuating entities and slower response for planning operators for planning purposes. The steady-state grid model
purposes) is typically composed of power-balance equations and equip-
4) practical alterations of grid model for scalable and effective ment limits. Power-balance equations dictate, for example,
design of signals (i.e., simplification of highly nonlinear grid how power flows through lines from generation sources to con-
constraints so that these capture important attributes of power sumers. Operational limits, in contrast, indicate the thresholds
balance while enabling tractable analysis) set for the safe operation of power equipment.
5) convergence of adaptive decision making (e.g., diversions Power balance is essentially determined by applying
that can lead to system damage) Kirchhoff’s voltage and current laws pertaining to steady-
6) computational and communication overhead associated state ac systems. The power system is composed of buses B
with designing and exchanging signals. to which consumers and/or generators connect. There are four
The proposed hierarchical framework will capitalize on important variables associated with each bus i ! B, and these
signal processing techniques to support the plug-and-play are bus-voltage magnitude Vi , bus angle i i, net real-power
integration of diverse power entities (e.g., renewables, flexible p i injection, and net reactive-power q i injection. This article
consumers, smart loads, and so on) across the grid; actuating uses capital letters to represent complex variables. Complex
nodes will be equipped with the ability to make preventative voltage Vi and apparent power S i can be expressed in rectan-
decisions that circumvent impending grid issues identified by gular form [9]
iterative signal and information exchange; system resilience
will be further strengthened by increasing stability margins via Vi = Vi cos (i i) + j Vi sin (i i), (1)
adaptive decision making facilitated by effective signal pro-
cessing; and efficiency will be embedded into every a­ ctuation S i = p i + jq i, (2)
decision made by intelligent nodes via adaptive construction
of monitoring and actuation signals. This article presents a where j = -1 . Buses are connected to one another by pow-
detailed exposition of various design aspects of the proposed er lines. Properties of line i - j connecting buses i and j are
hierarchical grid-management framework using mathematical encapsulated by the complex-valued line-admittance parameter
tools and techniques, such as convex optimization, multiagent
systems, machine learning, game theory, and social learning. Yij = g ij + jb ij, (3)
We include a detailed literature survey and practical simula-
tions along with theoretical studies to validate the effectiveness where g ij and b ij are the conductance and susceptance con-
of the proposed framework. stants. Complex power flow S ij from bus i to j is a function of
the bus voltages Vi, V j and the line admittance Yij
Tractable physical grid modeling
In the proposed framework, actuating nodes use signals to S ij = Vi (Vi - V j)* Y *ij . (4)
infer general trends in the system as well as local operating
conditions. To design signals that capture the grid state with For power balance to hold, the net complex power injected
respect to highest efficiency and feasible grid operating condi- by bus i must be equal to the power flowing from that bus to
tions, the incorporation of realistic physical power-grid mod- neighboring buses j as
els is imperative. As such, in this section, essential electri-
cal laws and limits governing power flow in the grid are first Si = / S ij = / Vi (Vi - V j)* Y *ij, (5)
presented. Then, challenges incurred in directly incorporating j ! Ni j ! Ni

86 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
where N i is a set representing buses that are directly connect- to heed steady-state physical grid requirements. Efficient
ed to bus i via a single power line (i.e., neighbors). Equation coordination in the power grid entails the construction of
(5) is a complex quadratic relation that can be equivalently ex- the optimization problem Pc
pressed in terms of real values as
Pc : min
x
f (x)
pi = / ` Vi g i, j - Vi V j g ij cos (i i - i j)
2 s.t. x ! S p,
j ! Ni

- Vi V j b ij sin (i i - i j) j which consists of the cost function f ; optimization variable x,


 which is a vector defined as { Vi , i i, p i, q i ; 6 i ! B}; and the
qi = / ` Vi 2 b i, j + Vi V j b ij cos (i i - i j) feasible set S p that imposes physical constraints on these vari-
j ! Ni
ables. The underlying characteristics of the objective function
- Vi V j g ij sin (i i - i j) j . (6)
and feasible set composing Pc dictate the level of difficulty ex-
pected in solving the problem. Specifically, the objective func-
It is evident that these power-balance equations are highly tion and feasible set must be convex for tractability [8]. Convex
nonlinear. functions are defined as
Each bus can be connected to a set of generators and/
or consumers. The net real- and reactive-power injection in if (x) + (1 - i) f (y) $ f (ix + (1 - i) y) 6 x, y

bus i is the cumulative contribution of these demand and ! dom f 6 i ! [0, 1] . (12)
supply units
Intuitively, this definition can be interpreted as forming a
g g
pi = pi - p di , q i = qi - q di , (7) linear line by connecting any two points x, y belonging to f
and all points on that line must lie above or on that function for
g g
where p i and q i represent total real- and reactive-power gen- convexity to hold. In the power system, the cost of generation
eration in bus i and p di and q di represent total real- and reac- and demand is typically a quadratic function that satisfies the
tive-power demand in bus i. Power generation and demand are definition of convexity [9]. A set Sc is convex if
subject to the upper and lower limits
ax + (1 - a) y ! Sc 6 x, y ! Sc, 6 a ! [0, 1] . (13)
g g -g g g -g
p-i # p i # p i , q-i # q i # q i . (8)
This definition can be intuitively interpreted as drawing
Flexible power consumers residing at bus i who have a line connecting any two points in the set, and all points
adjustable loads enable power demands to vary within the on this line must lie completely within the set. As nonlin-
thresholds ear equality constraints, such as (6), that compose the set
S p do not satisfy this definition of convexity, S p is not a
p-di # p di # p-i d , q-d i # q di # q-i d. (9) convex set. If the feasible set is not convex, then solving the
associated optimization problem will be NP-hard [10]. Thus,
Bus-voltage limits are also important considerations as optimal grid coordination that accounts for exact physical
these govern voltage stability constraints, such as power balance, becomes intractable very
rapidly. Applying decomposition to this nonconvex problem
v-i # Vi # vr i . (10) for designing signals will result in local actuations that lead
to system divergence and instability. Heuristic techniques
Voltage magnitudes are typically maintained around can be applied to solve these [11]. However, no guarantees
! 10% of the nominal value 1 per unit (p.u.) to prevent equip- with respect to convergence or optimality can be established.
ment damage. Apparent power limit reflects the maximum Instead, simplifications or relaxations can be carefully applied
2
square magnitude of complex power flow Sr ij permissible to these relations to render the feasibility set S p convex while
through a power line, and this is a constant parameter dic- retaining the defining characteristics of the physical grid.
tated by the characteristics of the associated conductor used This convexification process allows for the tractable compu-
for the line tation of the optimal solution with performance guarantees,
which, in turn, allows for the construction of effective moni-
2
p 2i + q 2i # Sr ij . (11) toring/control signals.

Every bus and line in the power grid is subject to con- Traditional grid operations
straints listed in (6)–(11). The state estimation of signals The traditional electric grid has been designed to accom-
generated by measurement devices, such as PMUs, will modate highly predictable power supply and demand. Thus
inform grid operators whether the system is operating far, grid operations have been well defined and mainly com-
within the physical grid limits. These constraints form the posed of contingency analysis, planning, and maintaining
feasible set S p to which the system variables must belong balance in demand/supply [3]. In contingency analysis, a

IEEE Signal Processing Magazine | May 2019 | 87


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
dynamic security assessment of the system is conducted where the grid. These mechanisms are based on highly simplified
offline and online simulations are run to ensure that the real-power-balance equations that neglect the contributions
grid operates within acceptable limits when various combi- of reactive-power and bus-voltage magnitudes. Other tertiary
nations of bulk synchronous generators are not operational control mechanisms, such as AGCs located in the governors
[12]. Specifically, N - 1 contingency is commonly upheld of synchronous generators, adjust mechanical power inputs
by grid operators to ensure that stability is maintained in via droop techniques (see the “Droop Control” section) to
the event that any one of the N genera- instantaneously adjust to changes in de­­
tion systems in the grid fails. In this case, Significant variability and mands inferred using local measurement
spinning reserves and ancillary services signals for maintaining nominal frequen-
wide margins of error in
are commissioned to increase the stability cies around acceptable thresholds.
margins during the recovery phase [13]. supply–demand forecast These grid-operation processes have
In planning operations, such as econom- models of modern power been highly effective until the recent pro-
ic dispatch, generation supply is matched entities prevent system liferation of renewable generation sources,
with forecasted demand in a cost-effective operators from predicting IPPs, and diverse power loads. Significant
manner ahead of time (e.g., a day ahead). well in advance reliable variability and wide margins of error in
Power suppliers and distributors partici- supply–demand forecast models of mod-
safety margins and an
pate in day-ahead markets to sell and pur- ern power entities prevent system operators
chase electricity [14]. Both the suppliers efficient balance of supply from predicting well in advance reliable
and distributors use day-ahead forecasts with demand. safety margins and an efficient balance of
of generation and demand for transacting supply with demand. Using AGCs for fre-
in the market. In Canadian provinces, such as Ontario, bids quency control is no longer sufficient, as this myopic tech-
are made, and an independent system operator will rank these nique cannot efficiently compensate for the considerable
bids and clear the market. The market-clearing price serves variability in power supply and demand introduced by renew-
as the cost signals used in decisions governing the buying/ ables and diverse power consumers. Moreover, since lines
selling power. No physical grid constraints are typically are operating close to established limits, reactive-power and
accounted for in this process [9]. Spot markets are based bus-voltage magnitudes are important considerations that
on similar principles but operate at a finer granularity (e.g., can no longer be ignored in power-balance equations. Thus,
hourly schedules). the highly granular coordination of active power entities by
The degree of balance in power demand and supply dic- designing signals based on realistic grid models is necessary
tates the system frequency in ac systems. Frequency control to overcome these limitations.
is maintained by generation sources at a much finer granular- However, as identified previously, incorporating exact
ity (i.e., seconds to minutes) via primary, secondary, and ter- physical grid relations, such as power-balance equations in
tiary control mechanisms [13]. Primary control mechanisms, the optimization formulation, leads to the nonconvexity of
which are typically employed in microgrid systems, focus on the feasible set S p . The distinguishing features of the system
eliminating adverse effects due to transient events. Second- being coordinated must be leveraged to relax these nonlinear
ary and tertiary control mechanisms adjust the operating set- relations into convex constraints. These defining character-
point signals of generation systems to correct for deviations istics are unique in four specific counterparts of the power
occurring over periods of minutes across various locations in grid: TNs, DNs, microgrids, and power consumers. In the
following, an overview is provided on how unique attributes
in each one of these systems can be leveraged to simplify S p
while preserving important system characteristics.

TNs
Bus
Power Line TNs transport power across geographically dispersed re-
gions that can span across hundreds of kilometers. All bus-
es in a TN operate at high voltages to minimize excessive
power losses that are typical in long lines. The ac power grid
employs three separate phases for increasing conductor effi-
Loop ciency and safety. As all three phases in a TN are balanced,
computations are based on single-phase constructs [3]. Thus,
the TN topology can be represented as a connected graph
where a single edge (power line) connects two nodes (buses).
Figure 5 presents the simplified graphical representation of a
TN consisting of 39 buses. A TN typically has a mesh struc-
FIGURE 5. A simplified graphical representation of the IEEE 39-bus New ture, as loops or cycles can exist in the power network, as
England system [15]. illustrated in Figure 5.

88 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
System operators typically apply approximations to TN lustrated in the sample 33-bus Danish DN system presented
power-balance equations based on three main assumptions: in Figure 6.
1) bus voltage magnitude does not deviate from the nomi- Each DN is managed by an EPU, which forecasts local
nal value Vi . 1 p.u.; 2) bus-angle differences are minor: demands and supplies bids to the system operator to pur-
i i - i j . 0; and 3) line admittance satisfies g ij % b ij . chase power from day-ahead bulk-electricity markets [20].
These lead to the dc power-flow equations that contain no The EPU bills power consumers using the advanced meter-
ac terms and no voltage and reactive-power terms [3] ing infrastructure, which consists of smart meters deployed
at each consumer unit. The smart meter is capable of bidi-
p i = / -b ij (i i - i j), q i = 0 6 i ! B. (14) rectional communication and actuation. Typically, industri-
j ! Ni al/commercial consumers participate in demand–response
programs, such as direct load control, where the EPU cur-
As the integration of unpredictable generation/loads threat- tails power consumed by these entities as necessary dur-
en voltage stability, cause significant uncertainties in power ing peak-demand periods via remote actuation signals to
actuation signals, and inflict congestions reduce stress on the system and provides
in the power lines [e.g., (10) and (11)], bus With the introduction of adequate compensation for these disrup-
voltage and reactive power flow cannot be EVs and other changes in tions [21]. Residential consumers partici-
ignored in the TN and, thus, the dc power- pate in the time-of-use program in which
the way consumers use
flow equations are no longer representative different electricity prices are allocated
of the TN.
power, DN bus voltages at on-peak, midpeak, and off-peak peri-
Many proposals in the existing literature can dip to unacceptable ods that serve as indirect signals reflect-
attempt to incorporate bus-voltage and values even during ing the congestion state of the grid [14].
reactive-power variables into the power- nonpeak periods. These give consumers the incentive to
balance model while ensuring convexity use less power during peak periods.
[16], [17]. One specific example is the use of a first-order With the introduction of EVs and other changes in the
Taylor series to approximate the quadratic terms in (6) way consumers use power, DN bus voltages can dip to
and a set of linear planes to estimate the cos (i i - i j) and unacceptable values even during nonpeak periods. Also,
sin (i i - i j) terms [18]. These changes transform the power- microgeneration sources deployed by consumers inject sur-
balance equations into a set of linear inequality constraints that plus generation back into the DN. This results in reverse
retain the bus-voltage magnitude and reactive-power variables. power flows that can lead to overvoltage conditions. These
The general form of the relaxed feasible set Slp with these under- and overvoltage conditions make voltage less stable
transformations in place is [i.e., (10)] and lead to equipment failure, the triggering of
protection devices, and, ultimately, cascading failures that
Slp = {Ax # B, Cx = D}, result in prolonged outages [4]. To prevent these calami-
ties, systems need active DN coordination accounting
where A ! R m # n, B ! R m, C ! R k # n, and D ! R k encapsu- for power balance and voltage stability by way of effec-
late constant parameters associated with the physical grid tive signal processing on grid-measurement data to reflect
and the related approximations. The general form of the these conditions.
relaxation is listed for the sake of brevity. These linear rela- The exact power balance and voltage-limit constraints list-
tions fairly accurately model power balance, as demonstrat- ed in (6)–(10) also apply to the DN. The simplification process
ed in [18]. These approximations hold only in TN systems, is, however, not the same as that of the TN, as the physical
not DNs, as assumptions, such as deviations in bus-voltage attributes of the DN are substantially different. In fact, the
magnitudes, are small and will not be applicable. Thus, three assumptions mentioned earlier obtain the dc power-flow
a different set of approximations must be applied in the equations in the TN do not hold for the DN. Much has been
DN setting. written about the relaxation of these constraints and about the
following two widely used models. The first is the linear Dist-
DNs Flow model. In this model, nonlinear power losses (i.e., I 2ij R)
DNs tap onto the bulk power supplied by the TN to pro- in DN lines are ignored. The resulting drop in voltage magni-
vide power to local residential, commercial, and industrial tude in line i - j can be modeled using the linear relation [22]
consumers at lower voltages. The connection between the
TN and a DN is established via a transformer substation v i - v j = 2 Re ^ Z ij S ijh, (15)
that steps down voltage from the TN side. Thus, there exists
2
a natural separation between the TN and DN. Power typi- where v i = Vi without loss of generality, the operator Re ($)
cally flows from the substation to individual consumers. As extracts the real component of the complex expression supple-
the DN operates at lower voltages, the voltage drops across mented as the argument and Z ij is the line impedance, which
these lines are not negligible. The physical network struc- is the inverse of line admittance. The complex power flow on
ture of the DN is radial (i.e., no loops), unlike the TN, as il- line i - j is then modeled as

IEEE Signal Processing Magazine | May 2019 | 89


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
| May 2019 | IEEE Signal Processing Magazine 90
B5
B3

H8 H4 H1 C1
B33

H9
H5
H10 C3
C2
H11 C5
H2
B1

B6

H3
Feeder 1

H6
H12
H7 C4
H13
B2

H14 C6
B4

H15
B9

H20
H21
H22 C9
H25 H18
B11

H26 H19
H16
B32

FIGURE 6. A graphical representation of the 33-bus Danish system [19].


C11 C10 C7
H27 H17
C12
H28 C8
H29 H23
B7

B10
B8

H24
B14

C13
Feeder 2

H30
B12

H32
H33 H31
C14
B31

H34 C16
C15
H35 H38
B34

B13

H36
H37
B16
Feeder 3

B15
H39
10 kV

10/0.4 kV

H40
H49 H41
H50 H42
H51 H43
C18

B22
C21
H52 H44
H54 H45
Substation

H53
B18
B30

H55 C17
H56 C22 H46 C19
H47

B23
H48
C23 C20
Feeder 4

H57 B17
H58

B25
B20
B19
H59
H62
H63 C24
H60
H64
H61

B21
H65
C25
H66
C26

B24
H67

Feeder 5
H68 C28
C27 H73
H69

B26
H74

B29

B28
H70 H75
H71

B27
H72
S ij = / S k, (16) power-balance relations. Capturing this transience in all three
k ! Ti phases present in the system using ac variables is not straight-
forward due to the nonlinearities inherent in these sinusoidal
where j is the parent of node i (i.e., located closer to the feed- variables. To overcome this difficulty, the dq0 frame of refer-
er), Ti represents the subtree rooted at node i, and S k is the ence is used to convert the three-phase sinusoidal ac system
complex power injection into bus k. This is a linear relation. states into 2D linear variables based on a rotating frame of
Neglecting power losses in the DN enables a conservative es- reference [25]. The transience caused by the inductance and
timation of voltage drop as these are strictly negative values in capacitance present in the lines and various power components
the power-balance equations. Relaxed feasible set has the form is captured in the resulting linear ordinary differential equa-
tions in the microgrid
Slp = {Ax # B, Cx = D}, (17)
xo dq = Ax dq + B, (20)
where A, B, C, and D are constant parameters.
The second model applies convex relaxations that transform where, for notational simplicity, constant coefficients of the
the optimal coordination problem into semidefinite (SD) and/ linear state variables x dq are grouped into matrices A ! R m # 2n
or second-order cone (SOC) programs [22]. The SOC A $ B and vector B ! R m, and these represent the microgrid attri-
relaxation converts the power-balance equations into convex butes. At the steady state, the differential terms will be 0, and
quadratic inequality constraints the resulting equations are linear in terms of x:

Slp = " Ax + B 2 # C T x + D, Ex # F, Gx = H ,, (18) Ax dq + B = 0. (21)

where A ! R m # n, B ! R m, C ! R n, D ! R, E ! R k # n, Thus, these linear equations are not only convex but also
F ! R k, G ! R l # n, and H ! R l are parameters representing exact and incorporate all three phases. For the conversion from
the relaxed physical grid attributes. SD relaxation introduces the three-phase abc to the dq0 domain, Park’s transformation
convex matrix inequality constraints in lieu of the original is applied. This is based on a common rotating frame of refer-
power-balance equality relations to obtain ence [3]. Every controller in the microgrid must maintain the
same frequency and phase for the rotating frame. These values
Slp = ) / A i x $ B, Cx # D, Ex = F 3, (19)
k
are generated by crystal oscillators present in these controllers
i =1 and are synchronized by GPS signals communicated between
these controllers [25].
where A $ 0 denotes that A is positive SD, and where A 2 B de- The dq0 frame of reference converts power-balance equations
notes that A - B $ 0, B ! R m # m and A 1 ! R m # m f A k ! R m # m into linear relations. However, when transforming other limits,
are matrices scaled by individual components of x. such as voltage-magnitude constraints, into the dq0 frame, these
These convex constraints obtained using SOC and SD become nonconvex relations. To overcome issues pertaining to
relaxations are computationally more involved than the linear nonconvexity, linear approximations can be applied in a manner
DistFlow equations. However, one major advantage is that, similar to [18] to obtain
under certain conditions (e.g., radial structure and others) out-
lined in [23], these relaxations are exact in the DN. Thus, the Slp = {Ax dq = B, Cx dq # D}. (22)
tradeoffs between the linear DistFlow model and SD/SOC
relaxations are the tightness of the relaxation versus computa- Power consumers
tional efficiency. Power consumers are increasingly deploying storage sys-
tems, microgeneration systems (e.g., solar panels), and smart
Microgrids loads (e.g., EVs) in their premises, and these can be effectively
Microgrids contain generation sources and consumers. Like the coordinated by smart energy-management systems (EMSs)
DN, these directly supply electricity to consumers and are typi- [26]. Storage devices and DG systems can be combined to
cally deployed in remote communities. Microgrids can operate reduce the amount of electricity purchased from the main
in grid-connected or islanded mode. In grid-connected mode, grid. This promotes energy independence. When it is neces-
perturbations in demand/supply are absorbed by the main grid. sary to purchase power from the main grid or to conserve
In islanded mode, the microgrid, unlike the DN, is not con- energy, smart appliances can be ­coordinated accordingly to
nected to the main grid, and therefore, the inertia maintained ­maximize c­ onsumer comforts. The d­ efinition of user comfort
by bulk-generation sources is no longer present. The microgrid is unique to each consumer and can vary based on ­diurnal pat-
will then need to be self-sufficient and use fast-acting signal terns, seasons, and weather conditions. User comfort is associ-
processing and control mechanisms to stabilize the system dur- ated with the tolerance Pr rj to change at various c­ oordination
ing transient periods [24]. ­horizons [27]
As transience can seriously undermine the stable opera-
tion of the microgrid, it is necessary to account for these in the 0 # P rj # Pr rj, (23)

IEEE Signal Processing Magazine | May 2019 | 91


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
where P rj represents the reduction in power demand by consum- Decision-making topologies
er j. Energy budgets E dj representing the maximum amount of Decision-making topologies govern the way actuating power
energy reduction tolerated by the consumer will ensure that en- entities exchange information and use externally/locally gen-
ergy is not cut too much: erated signals to determine choices. These intelligent actua-
tion processes must foster greater efficiency in the system
E rj + DtP rj # E dj , (24) while heeding physical infrastructure constraints. Thus, key
elements of the decision-making process are 1) the effec-
where E rj represents energy reductions so far during the day, tive construction of signals that succinctly capture relevant
and Dt is the length of the current coordination interval. information about the operating status of the system; 2) the
Power reduction is typically discrete as appliances function at manner in which signals are exchanged among participating
specific power levels or operate in a binary fashion (i.e., on/ nodes; and 3) effective processing of these signals for fea-
off). Hence, P rj ! D j, where D j represents a discrete set of sible and efficient decisions. Four different types of decision-
power-reduction values. These are combined to construct the making topologies are presented in this section, and these are
feasible set based on centralized, decentralized, distributed, and inde-
pendent coordination paradigms. The suitability of each one
S p = " Ax = B, Cx # E, x ! D ,, (25) of these coordination constructs depends on the application
(e.g., planning, preventative actions, efficient power balanc-
where A, B, C, and E are constant parameters. Discrete ing, and others) and system characteristics (e.g., TN, DN, mi-
variables introduce nonconvexity. However, game theoretic crogrid, and others).
­constructs can be applied to overcome is-
sues due to discreteness, as is discussed Decision-making Centralized coordination
later in this article. topologies govern the way Centralized coordination consists of a single
management entity (e.g., a system operator)
actuating power entities
Summary of physical grid models that computes the proper operation setpoint
Realistically modeling the underlying elec-
exchange information signals for each actuating entity present in
trical characteristics of the power grid is and use externally/locally the system [3]. To incorporate physical in-
a vital step in building an efficient grid-co- generated signals to frastructure limits into the computations,
ordination framework based on signal determine choices. the management entity must be aware of
processing. The grid model must allow for detailed nuances associated with the physi-
tractable computations while maintaining important physi- cal topology (e.g., connection structure of nodes, node/line at-
cal attributes of the system. The simplification processes tributes, actuation limits, and others) of the entire system. This
presented for the four main components of the power grid significantly limits flexibility in the system, as the central con-
have resulted in the system models summarized in Table 1. troller must be aware of every change taking place through-
Usually, in the selection of the underlying grid model, the out coordination domain. To centrally solve the coordination
main tradeoffs are between computational complexity and problem in a tractable manner, appropriate relaxations depend-
tightness of the relaxations (i.e., representation of the original ing on the system being coordinated can be applied to the fea-
constraint set). sible set S p, as discussed in the previous section, to render the
Now that the physical grid constraints can be incorporated Pc convex. Then, well-known convex solvers can be used to
in a tractable manner into the optimization framework, the solve the relaxed problem in polynomial time O^n p h, where n
next step will be to explore how signals and actuation can be represents the number of entities being coordinated [8].
designed to enable intelligent actuating nodes to adaptively Extremely powerful computational resources, such as
make the best actuation decisions. cloud-computing platforms, can be used to centrally compute
the best solution to the convex coordination problem [28]. The
main issues with this setup are threefold: 1) highly fluctuat-
ing grid parameters; 2) the risk of data exposure; and 3) the
Table 1. A summary of physical power-balance constraints.
single point of failure. Since grid parameters, such as genera-
System Model of Power Balance Complexity of Terms Tightness tion capacities and demand patterns, cannot be accurately pre-
dicted using long-term models, the central coordinator must
TNs dc power flow Linear Weak
Linear ac approximations Linear Tight
accrue these values directly from the varying nodes over short
DNs Linear DistFlow Linear Tight intervals [29]. These data signals sent by individual nodes
SOC relaxations Quadratic Exact will result in the establishment of n dedicated communica-
SD relaxations Positive SD Exact tion links with the central controller. This information must
Microgrids ac power flow Sinusoidal Exact be processed to compute the optimal setpoints, which are then
dq0 frame Linear Exact
conveyed individually to every actuating node by the coor-
Power Comfort requirements Linear Exact
consumers Appliance actuation Discrete Exact dinator. This necessitates the forging of n more dedicated
communication links in the reverse direction. Hence, at every

92 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
coordination period, 2n point-to-point links are established to
Table 2. A summary of overheads in centralized coordination.
exchange measurement and actuation signals. When there are
thousands of highly varying nodes in the system (e.g., TN or Complexity
DN), the ensuing overheads become excessive, as summarized System Nodes n C
­ omputational Communication
in Table 2, where k, k 1, k 2, and p are constants. TNs 1,000 p
O (n ) . k 1 1, 000 p
O (n 2) . k 2 1, 000 2
Moreover, these frequently exchanged signals processed DNs 1,000 O (n p) . k 1 1, 000 p O (n 2) . k 2 1, 000 2
at a central location can be easily intercepted via cyberchan- Microgrids 10–100 O (n p) . k 1 100 p O (n 2) . k 2 100 2
nels. In this case, highly revealing information about the grid Power consumers 10–20 O (k n) . k 20
1 O (n 2) . k 2 20 2
and its operating trends can be gleaned by an adversary and be
exploited to perpetrate an insidious cyberphysical attack on the
system [30]. To address these security issues, additional lay- demand and supply in the system. The design of these signals
ers of encryption can be added to the data being exchanged. is illustrated via two different approaches: 1) the subgradient
This can, however, impose greater overheads to the already (SG) method and 2) the water-filling technique. Every actuat-
resource-intensive centralized coordination process. Also, if ing node i (e.g., storage systems, DGs, and others) has its own
the central entity coordinating the entire system is not func- definition of cost fi (x) and, hence, the objective of the coordi-
tional, then this single point of failure can undermine the oper- nation problem is separable. Power-generation capacities and
ation of the entire system. other constraints particular to each active node i form the local
Central coordination is not suitable for controlling a vast feasible set Si . The only binding constraint is the global bal-
system with many nodes. As it is not very scalable, central ance in demand and supply:
coordination is more suited in a smaller and contained setting,
n
such as for a home or building or over long control horizons. PED : min
x i ! Si
/ fi (x i)
i =1
Distributed coordination n n

In distributed coordination, actuation decisions are made lo-


/ A TI x i - /
A TI x i = 0,
i!G i!D
cally by every active node instead of by a central entity. These
decisions depend on external signals communicated periodi- where x i represents the optimization variables pertaining to node
cally. Unlike decentralized coordination, the distributed coor- i, and the sets G and D represent nodes that generate power
dination process involves a central entity that provides general and consume power, respectively. The variable set in D remains
information to actuating nodes in the form constant in economic dispatch problems as
of broadcast signals computed using aggre- Extremely powerful the generation setpoints of variables in G are
gate monitoring data encapsulating global computational resources, optimized based on the associated cost and
system trends. overall demand in the system. When demand
such as cloud-computing
For example, in the DN, the EPU is a cen- response is considered, power consumption
tral entity that manages all of the electricity platforms, can be used to by flexible consumers in D is adjusted to
billing processes and, thus, has access to centrally compute the best meet the overall power injected into the sys-
information generated by smart meters [20]. solution to the convex tem via generation sources. The value A I
This information can be aggregated by data coordination problem. serves as an indicator vector for notational
concentrators in a manner that reveals the convenience that selects the real power p i
general state of the DN with respect to voltage stability, load variable from the set x i . It is assumed that both the demand and
balancing, costs incurred, sustainable power consumption/ supply variables are continuous. The dual problem is then con-
generation, and so on. General trends inferred from this infor- structed by formulating the Lagrangian in which the dual vari-
mation are then broadcast by the EPU to individual actuating able o associated with the binding constraint is introduced [8]:
elements (e.g., power consumers, micro-DGs, storage systems,
: max min / fi (x i) - o e / A TI x i - / A TI x i o .
n n n
and so on), which can then respond adaptively to increase D
P ED
system efficiency while heeding local constraints. The EPU v ! R xi ! Si
i =1 i!G i!D
does not need to form dedicated point-to-point communication
links, as the information broadcast is common to all entities In the SG method, the dual problem is decomposed by
in the system. The issue of concentrated computation is also grouping optimization variables associated with each actuat-
resolved, as this is offloaded to individual actuating agents that ing node together to form the local optimization problem [31]:
act in parallel. Two approaches based on dual decomposition
and population game (PG) theory illustrate the effectiveness of P SED : min fi ^ x ih - o ^I G (i) A TI x i - I D (i) A TI x i h,
xi ! Si
the distributed coordination paradigm.
where I G (i) is an indicator function that returns one if node
Dual decomposition i is a generation source and zero otherwise. The value I D (i),
In the first approach, signals constructed by the central aggre- defined in a similar manner, is used to identify whether node
gating entity aim to establish a balance in the overall power i is a power consumer. The variables o and x are computed

IEEE Signal Processing Magazine | May 2019 | 93


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
iteratively by the central aggregating entity and individual Si . Since PED is convex, these iterative revisions are guaran-
nodes. The central aggregating entity computes o by fixing x teed to converge [33]. However, the step-size a must be cus-
D
to the current value taken in the system at time t in P ED . This tomized for the system under consideration to avoid oscillatory
value o is broadcast to individual agents, which then substitute behavior, as illustrated in Figure 7, where the total demand is
this into the variable o to compute x i by solving P SED . fixed and local generation systems adaptively actuate the real-
Thus, the original problem is divided into master and slave power injection to match this demand. The coordination occurs
problems where, in the master problem, the central coordinat- during every 10-min interval, and the system consists of 20
ing entity attempts to iteratively compute o via the SG update large-scale renewable sources with highly fluctuating genera-
technique, where q (o) is the SG of the master problem at point tion capacities. Clearly, these individual sources are able to rap-
o when x i is fixed and a is the step size [32]: idly converge to the best solution for the appropriate step size.
D
Alternatively, P ED can also be solved using the water-fill-
o t +1 = o t + aq (o t) . (26) ing method [34], where the central coordinating entity com-
putes o iteratively using binary search and the stopping criteria
The SG turns out to be the difference in aggregate genera- / di = 1 x di - / gi = 1 x gi = 0. Individual agents, on the other hand,
tion and demand at time t: receive o t broadcast by the aggregator and analytically com-
pute local actuation by applying the Karush Kuhn Tucker
n n
optimality conditions, which are necessary and sufficient in
q (o t) = / A TI x ti - / A TI x ti, (27)
i!D i!G convex optimization problems [8].
First-order optimality is expressed as
2L ^ x *, o * h
and this is an aggregate measure that can be easily obtained
from data concentrators. This value is broadcast by the central = 0.
2x i
aggregating entity in a periodic manner (e.g., every second).
In the slave problem, each actuating node computes the opti- Primal and dual feasibility is expressed as
mal value of x i, which is a function of o t and local constraints
x * ! S p, o * ! R.

Complementary slackness is expressed as


× 104 Matching Supply and Demand
o ^ I G (i) A I x i - I D (i) A I x i h = 0,
12 * T * T *

10 D
where L is the objective of the dual problem P ED , and x * and
*
8 o are optimal values of the problem. This method is called the
Power (kW)

water-filling method, as the local constraints, such as genera-


6 tion capacity c i, form a boundary analogous to an enclosure,
as illustrated in Figure 8. The water level represented by -o
4 Total Generation (α = 1) reflects the aggregate state of the system and, this is increased
Total Generation (α = 2) until the stopping criteria is met (i.e., demand is equal to sup-
2 Total Generation (α = 10) ply), at which point, the optimal solution x * is achieved. Like
Total Demand
the SG method, the convergence of the system to this optimal
0 point depends on how o is updated. With the binary search
0 0.1 0.2 0.3 0.4 0.5
method, oscillations will be present. Also, the convergence rate
Time (h)
of both methods is proportional to the number of nodes in the
FIGURE 7. The impact of a on convergence using the SG method. system O (n) [33]. Thus, the three main issues associated with
the SG and water-filling methods are as follows: 1) actuation is
based on a continuous domain (e.g., smart appliances operate
in discrete power levels); 2) smooth convergence entails fine-
Demonstration of Water-Filling for Larger –v tuning of the updating parameters; and 3) convergence time
Cn
depends on the number of nodes being actuated.

xn
PG theoretic approach
–v Introducing discrete variables can render Pc an NP-hard prob-
lem [10]. When there are thousands of discrete variables in-
volved (e.g., in DNs with thousands of appliances), solving the
x problem as is becomes impractical, as NP-hard problems are
not scalable or tractable. Instead of directly solving Pc , apply-
FIGURE 8. A water-filling analogy [34]. ing a PG theoretic approach to transform this problem into a

94 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
Min and Max Voltage Profile Over a Day With Dispatch
Comparison Between PG and SG Methods
1,000
1.1
PG
Setpoint
800
SG: α = 0.0001
Aggregate Power (kW)

Normalized Voltage
1.05
SG: α = 0.0024
600
1
400
0.95 Max Voltage
200 Min Voltage
Voltage Limits
0.9
0
0 10 20 30 40 0 5 10 15 20
Iterations Time (h)

FIGURE 10. The bus-voltage magnitudes with a feasibility check. Min:


FIGURE 9. A comparison between the PG and SG methods. minimum; Max: maximum.

Z
game consisting of thousands of players equipped with discrete the distribution Z in effect is the optimal solution for P ED
strategy sets resolves many of the previously mentioned issues, [35]. This can be proven by showing that the system dynam-
as discussed later in this section [35]. The main premise behind ics ensuing from these incremental revisions has a Lyapu-
the PG approach is that, when one player switches strategies, nov function, which is exactly L ^o, Z h . The existence of a
the effect on the overall system cost is incremental. If individual Lyapunov function guarantees convergence to Z *, which is
strategy revisions are made so that the overall potential/cost of the point that results in the gradient of the Lyapunov func-
the system decreases over time, then an equilibrium state will tion being zero (i.e., 2L /2z = 0) [35]. This is also the condi-
be reached eventually [36]. tion for first-order optimality, which implies that Z * is the
Z
All possible discrete strategies available in the system form solution of P ED . The convergence speed of this method is
the set Y = " y 1, f, y m ,, where y i represents the discrete independent of the size of the system, as aggregate measures
power level associated with strategy i. The central aggregating compose the variable set.
entity maintains another variable Z = " z 1, f, z m ,, where z i is Figure 9 compares the convergence of a system composed of
an aggregate variable that represents the fraction of agents in 1,000 nodes representing smart appliances over a single coor-
the population that are using strategy i. Thus, Z is a continu- dination period of 1 min via the PG and SG methods. Realistic
ous variable defined over a convex simplex [37]: demand models, appliance usage statistics, and penetration rates
have been employed for these simulations. At each iteration, the
m
Z = 'z ! Rm ; / z i = 1, z i $ 0 6 i ! m 1 .(28) central aggregating entity broadcasts a signal to the actuating
i =1 nodes. With the SG method, it is assumed that the actuation
(i.e., demand curtailment) is continuous. With the PG theoretic
D
This variable substitution can be applied to P ED , and this method, more realistic discrete strategies are used instead. The
Z
transformed problem is referred to as P ED with an objective PG theoretic method displays fast convergence and exhibits
L ^o, Z h . The central coordinating entity can now directly no oscillations. This is not the case with the SG method when
solve this dual problem to obtain o * and Z * . The main chal- a = 0.0024. Although the PG theoretic method is highly effec-
lenge now lies in achieving this optimal distribution Z * in tive for large-scale discrete coordination, it cannot be applied
the population. Every agent may have different local operat- to systems consisting of a few nodes, as the stochastic nature of
ing conditions and constraints that the central coordinating the random decision-making process will take effect and introduce
entity does not yet know. Thus, instead of directly actuat-
ing individual agents, the coordinator computes the gradient
Z
of P ED with respect to each strategy to obtain 2L /2z and Table 3. A summary of distribution decision-making topology.
broadcasts this gradient vector to all agents in the system.
Coordination Component System Convergence
The agents will revise their local actuation strategy at a
randomly selected time in a manner that accounts for local Aggregating signal SG O (n)
constraints and moves the aggregate state of the system in Water filling O (n)
PG O( Y )
the opposite direction of the gradient. This process will be
Local feasibility check Tree network O (d )
repeated until there is no incentive to switch (e.g., gradi- Mesh network O (n)
ent is 0). At this point, Nash equilibrium is achieved, and

IEEE Signal Processing Magazine | May 2019 | 95


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
perturbations. This is eliminated in a system consisting of many limits, the revising agent will not proceed with the computed
participants due to the strong law of large numbers. change in actuation.

Local feasibility checking Summary of distributed coordination


So far, three methods have been presented for designing sig- In the distributed coordination technique, a central aggregating
nals broadcast by the central aggregating entity in the distrib- entity informs participating nodes of the global trends in the sys-
uted setting. Active nodes factor these signals into selecting ap- tem via periodic broadcasts of the generalized signals. Agents use
propriate local actuation. Prior to implementing the computed these signals to make local strategy revisions while also account-
actuation, the node proceeds to evaluate whether or not the ing for local infrastructure limits. These signals are designed us-
actuation strategy at hand violates local capacity limits and/or ing aggregate measures (e.g., a surplus/deficit in demand/supply,
infrastructure limits. Local capacity limits are embedded into the proportion of agents using a particular solution, and so on),
the computation of the actuation signal via the SG and water- which are readily available via data concentrators and supervisor-
filling methods. With the PG method, discrete strategies that control and data-acquisition systems. Convergence properties are
violate local capacity limits are not considered as candidates analyzed using convex optimization techniques (e.g., SG and wa-
for strategy revision. Infrastructure limits, such as bus-voltage ter-filling methods) and control theory (e.g., Lyapunov functions).
and apparent line limits, are taken into account by way of com- Table 3 presents a summary of the performance characteristics of
munication with neighboring nodes. these distributed decision-making systems.
In the DN or microgrid, which has a radial topology, an This distributed coordination paradigm naturally fits into
actuating agent selects a random time for making the strat- systems, such as DNs, as there is a central entity, such as an
egy revision so that its strategy change does not coincide with EPU, that has readily available access to aggregate measure-
other nodes in the system. At this time, this node commu- ments in the system. Moreover, the radial structure of the DN
nicates with bus agents residing in its local feeder branch to allows for localized checks of physical power-balance con-
ascertain that the impending change in power flow will not straints. However, this technique will not be appropriate in
violate bus-voltage magnitudes or apparent power-flow lim- deregulated systems, such as TNs consisting of independent
its (e.g., using DistFlow equations) [38]. The communication entities (e.g., IPPs) that are not centrally managed. In these
complexity of this process is O (d ), where d is the depth of cases, a completely decentralized solution where no central
the tree representing the DN rooted at the substation. Fig- aggregating entity is present is suited for establishing coordina-
ure 10 illustrates the maximum and minimum voltages in the tion among these independent elements.
33-bus DN with this local feasibility check in place for the
PG-based distributed coordination [19] composed of 1,000 Decentralized coordination
active nodes. Bus-voltage magnitudes are within acceptable In decentralized coordination, individual actuating entities itera-
limits of ! 10%. tively exchange signals with nearby nodes to make local deci-
In the TN, which has a mesh topology, this can be verified sions. As no central entity is involved in directly actuating or co-
via repeated exchange of local state signals with neighboring ordinating nodes, no single node needs to be aware of the entire
nodes. Specifically, the alternating direction method of multi- physical network structure or needs to gain access to aggregate
pliers (ADMMs) can be used to evaluate whether i­ nfrastructure data sets. This aligns with the deregulated nature of the modern
constraints are violated, as discussed in the section “Decentral- power grid and eliminates single-point-of-failure issues. In the
ized Coordination.” The complexity of this process is O (n) presence of malfunctioning power devices, normally operating
[39]. If the impending strategy revision violates infrastructure nodes can infer these anomalies via signals exchanged among
each other and adaptively modify local actuation to isolate these
issues and reestablish the nominal operation of the system. In the
Grid Monitoring Based on Social Networking
power engineering literature, the three common coordination
0.92 techniques for the decentralized coordination of demand and/or
supply are the consensus method, ADMM, and potential games.
0.91
Extensions of these techniques are used in applications related
0.9 to grid monitoring, power balance, and preventative actuation.
V_min

0.89 Voltage Threshold


Consensus method
0.88 In the consensus method, all participating agents repeatedly
No Monitoring exchange information with one another to reach an agreement
0.87
Opinion Dynamics
(e.g., total demand is equal to total supply). Convergence to the
0.86
0.625 1.25 1.875 2.5 3.125 3.75 4.375 5 average consensus is proportional to the largest eigenvalue of
Total EV Load (MWh) the Laplacian matrix representing the information exchange
network topology [40]. This notion of consensus is also used
FIGURE 11. The integration of EVs with monitoring based on social for inferring trends in social networking platforms, which can
networking. be applied in the grid context for monitoring processes [41].

96 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
Every node in the power grid can have an opinion s i (t) about i
P ADMM: min fi ^ x ih
x i ! X i, y i ! Yi
the operating state of the power grid at time t. This opinion
can be exchanged repeatedly among immediate neighbors un- x j = y ij 6 j ! N i .
til an equilibrium is attained. DeGroot’s method, which en-
tails the computation of myopic averages by individual nodes, The third set of variables o ij are the Lagrangian mul-
is typically applied in this context. As such, the evolution of tipliers associated with the consensus constraint between
i
opinions (i.e. opinion dynamics) over time via this method can the actual and perspective variables as listed in P ADMM .
be modeled as The goal of every node will be to attain consensus (i.e.,
x j - y ij = 0 6 j ! N i) between the perspective y and actual
s (t) = A d s (t - 1), (29) x variables where o ij indicates the degree of mismatch.
This can be achieved by first constructing the aug-
where A d is the adjacency matrix representing the under- mented Lagrangian L t (x, y, o) of PADMM that contains an
lying topology of the communication network, and s (t) = additional term squaring the mismatch between the per-
" s 1 (t) gs n (t) , is the vector containing the opinion of all spective and actual variables. This term is weighted by a
n nodes at discrete time steps t. This refinement of opin- positive constant t. Then, the iterative ADMM updating
ions via DeGroot’s method will eventually converge to the technique is applied to the associated variables as [42]:
global average of the opinions in the network as long as the
communication network is fully connected. This global av- x k +1 = argmin L t ^ x, y k, o k h,
x

y k +1 = argmin L t ^ x k +1, y, o k h,
erage can be used to deduce the state of the network (e.g.,
congested, stressed, healthy, and so on), which enables indi- x
vidual agents to adaptively respond to improve the general o
k +1
= o k + t ^ x k +1 - y k +1 h .
health of the system.
In Figure 11, this social networking method has been This process entails the exchange of three different sets
applied to determine the connection of EVs to the grid for of parameters with neighboring nodes at each updating
charging purposes in a 128-bus DN system. With this decen- iteration k. These sets of parameters are guaranteed to
tralized monitoring in place, it is clear that dangerous voltage converge if Pc is convex. This is indeed the case when
violations due to congestion can be prevented. convex relaxations are applied to the power-balance rela-
tions. Furthermore, the convergence rate is proportional
ADMM approach to the number of nodes n in the system [i.e., O (n)] . Fig-
Establishing optimal power flow via the ADMM approach en- ure 12 shows the ADMM method applied in the TN set-
tails the design of signals that incorporate information about ting using the linear ac approximation of power-balance
local infrastructure states (e.g., the apparent power-flow and constraints and the change in the residual ^i.e., x - y 2h
bus-voltage magnitude) by every actuating node. These sig- over one coordination interval of 10 min. Signals are
nals are then exchanged with the nodes of direct neighbors exchanged every 6 s. This system is composed of 2,736
and used to refine local actuation to increase efficiency while buses [15], and it is clear from Figure 12 that the decen-
maintaining locally inferred feasibility. Signals are repeatedly tralized coordination mechanism results in fast conver-
exchanged with peers until a global consensus regarding the gence to the optimal solution. Thus, the communication
nominal operation of the entire system is attained. of signals to neighboring nodes that contain current val-
The design of signals is composed of two steps. First, the ues computed for local, perspective, and dual variables
nonconvex power-balance relations are converted into a set of enables every node to adaptively respond to changes in
convex constraints via one of the linear ac approximations, the system in a decentralized manner while accounting
convex relations via SD/SOC, or steady-state dq0 transforma- for physical power-balance relations.
tion, as listed in the section “Tractable Physical Grid Model-
ing.” The next step will be to decompose these power-balance
equations for each node to infer local feasibility. However, as Residual at Each ADMM Iteration
power balance depends on power flowing from directly con- 20
nected nodes, it is not possible to directly decompose this 15
Residual

for individual buses. To render the power-balance equations


separable, each node maintains three different sets of vari- 10
ables [39]. The first set x i contains local variables pertain- 5
ing to the actuating node (e.g., local generation, local voltage
magnitude, and so on). These variables are subject to local 0
20 40 60 80 100
constraints X i . The second set of variables are the perspec- Iterations
tives y ij maintained by node i of variables belonging to all
neighboring nodes j ! N i, which are subject to constraints FIGURE 12. Optimal power-flow coordination via ADMM in the Polish
i
Yi . These constitute the local problem P ADMM : 2,736-bus system.

IEEE Signal Processing Magazine | May 2019 | 97


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
Potential game approach Nash equilibrium in which no agents shall regret their current
Decentralized coordination can also be applied in preven- state (i.e., it is not possible to deviate further without incurring
tative measures where build-up of congestion/stress can be additional costs). These repeated revisions of discrete strate-
detected and individual agents can actuate to avoid poten- gies will result in the optimal topology configuration if dis-
tial cascading failures. For instance, the physical network crete concavity conditions are met [44]:
topology in the DN is typically fairly static. Individual
2 min (U (x), U (y)), U (x) ! U (y)
U (z) = '
switches can be activated to add or remove lines to re-
max ,
configure the system topology in case of line outages and $ U (x) = U (y), U (x) = U (y)
x, y, z ! X: x - z = y - z = 1
other emergencies. These switches are typically reconfig- (31)
ured in the aftermath of an event not a priori. With the
proliferation of EVs and highly variable renewable loads, where X is a set consisting of all configurations of individual
loads across the DN feeders can be unbalanced and volt- switches and x, y, and z are vectors that belong to this set X
age drops across lines can be excessive. The bus switches in which the ith element is zero if switch i is connected to
can be activated to achieve a more efficient topology and the original node and one if it is connected to the new node.
reduce the risk of failure The value U is the utility function that depends on the cur-
This topology reconfiguration problem consists of discrete rent state of switches in the system. Typically, heuristic algo-
variables (e.g., a switch or no switch) that render the problem rithms are used for this purpose, and these have no guarantees
very difficult to solve directly. A decentralized algorithm based on convergence and optimality. Figure 13 illustrates the ef-
on the potential game approach enables every bus switch to fectiveness of the potential game approach for DN topology
decide whether to shift to a new node by communicating with reconfiguration in comparison to the heuristic fast nondomi-
neighboring peers and, thereby, identifying the impact of the nated sorting genetic algorithm (FNSGA) [45]. In this figure,
switch on power balance (e.g., modeled using linear DistFlow I indicates how close the result from applying the algorithms
equations) in the network. These decisions are sequential, and is compared to the optimal voltage profile computed using the
the topology resulting from a switch at time t is denoted as Tt . brute-force method, which applies load-flow analysis using
These sequential decisions improve the voltage profile V ^Tt h ­MatPower. On the x axis, the tuple 69 - 5, for example,
of the network [43]: models a 69-bus DN consisting of five switches. Clearly, the
potential game approach yields far better results than the heu-
V ^T0h 2 V ^T1h 2 V ^T2 h 2 g(30) ristic technique.

As the set of strategies available to each switching agent Summary of decentralized coordination
is finite, this sequence will converge within finite time to a Decentralized coordination enables individual agents, by ex-
changing signals with neighboring nodes, to adaptively re-
spond to changes and perturbations and, thereby, progressively
improve the global state of the system while adhering to local
Confidence Interval of Improvement Factor constraints. Table 4 summarizes the three decentralized meth-
1.05
ods and applications for adaptive and proactive actuation that
1 have been presented.
0.95
0.9
Independent coordination
In all of the coordination paradigms presented earlier, the
0.85 nodes communicate. The communication channel is subject to
Potential Game
0.8
FNSGA
latencies in the order of milliseconds [46]. In certain applica-
0.75 tions that require fast-acting control and response, this latency
33-3 33-5 33-7 69-5 will not be tolerable (e.g., primary control) [24]. In these sce-
DNs narios, only local measurements (e.g., bus-voltage magnitude,
current, frequency, and so on) can be used by actuating nodes
FIGURE 13. A comparison of the performance of DN topology reconfigura- for inferring the general state of the system, and actuation is,
tion algorithms. therefore, independent of external input. Specific actuation
mechanisms include those based on droop control, primal-dual
dynamics, and machine-learning techniques.
Table 4. A summary of decentralized algorithms and applications.

Method Application Convergence Droop control


Droop control is widely used in both TN and microgrid set-
Consensus Grid monitoring O (n)
tings by generation systems that measure local frequency de-
ADMM Power balance O (n)
Potential game DN topology reconfiguration O (k ) viations to infer the degree of mismatch in demand/supply in
the system [3]. If there is higher power demand in comparison to

98 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
the supply in the system, generation frequency will decrease age s­ tability. Voltage magnitude Vb of bus b must lie within
and, thus, more real power must be put out by the generator the 1 ! 0.1 p.u. threshold for the nonfaulty operation of power
and vice versa when the demand is lower than supply. To components. At the extreme ends of this spectrum, when Vb
maintain the frequency of the system around nominal val- is 1.1 p.u., generation is excessive in comparison to demand
ues, the generator controller uses the relation [3]: and, when Vb is 0.9 p.u., demand is much higher than gen-
eration. The work in [48] divides the interval 60.9, 1.1@ into
~ = ~ ref - k p (P - Pref ), (32) subintervals, where each subinterval represents an operation
mode. Based on the operation mode at hand, actuating grid
where k p is the droop coefficient constant associated with elements, such as storage systems, DGs, and loads, actuate at
real load sharing, and ~ ref and Pref are the reference frequen- three modes: maximum power draw/injection, voltage con-
cy and reference active power setpoint computed by solving trol, minimum power draw/injection. These operational and
an economic dispatch problem, respectively. If the locally actuation modes are derived via machine-learning concepts
measured frequency decreases to ~, then active power P is such as fuzzy-logic techniques. Other adaptive controllers,
increased to the value obtained by solving (32). This droop such as those using model predictive control techniques, are
control mechanism allows for active power sharing among also being proposed for independent actuation using local
multiple generation units in the event of minor disturbanc- measurements [49]. Although machine-learning techniques
es or deviations of demand/supply from forecasted values. allow for the adaptive configuration of controllers for main-
These myopic actuation decisions are especially suited for taining feasibility in infrastructure operation, optimality or
systems that do not experience significant deviations from efficiency cannot be guaranteed due to the lack of coordina-
forecasted values. However, with the recent proliferation of tion among other nodes in the system and empirically derived
renewables and high-power consuming devices, these myopic control parameters.
decisions will not be efficient, as these will attempt to main-
tain setpoints that are unrepresentative of the actual condi- Summary of independent coordination
tions in the system. Independent coordination involves no communication, and
decisions are made using the local measurements of system
Primal-dual dynamics states. This myopic technique is simple and not subject to un-
To ensure that actuation computed using local measurements expected delays or vulnerabilities in the cyberchannel. How-
is optimal in a simplified context, recent work on primal-dual ever, achieving guarantees on efficiency and optimality while
dynamics capitalizes on problem formulations, such as PED, to accounting for all-important state variables, such as reactive/
use local frequency and power-flow measurements to iterative- real power-flow and bus-voltage magnitude/frequencies with
ly compute the optimal actuation for addressing this problem. no access to the global system state, is not practical, as sum-
First, a dual variable o for the overall balance in power de- marized in Table 5. Thus, independent coordination is highly
mand and supply is introduced. This variable o i is maintained suited for such applications as primary control, where response
locally by every node. At optimality, o i maintained by node i time is critical, and not for optimal operations.
must be equal to the oj maintained by node j. Another dual
variable r ij is introduced to represent this synchronization of Integrated hierarchical decision making
o among nodes i and j. The work in [47] proved the equiva- The power grid is a highly coupled and complex system com-
lence of o i with local frequency ~ i and r ij with power flow posed of many diverse and fluctuating power components. We
Pij from node i to j based on the simplified linear dynamical have presented a detailed exposition on the tractable modeling
equation of power flow: of the electrical characteristics/limits of the existing grid in-
frastructure, associated system stakeholders (e.g., TNs, DNs,
DPo ij = B ij (D~ i - D~ j), (33) microgrids, power consumers), and decision-making para-
digms (i.e. centralized, distributed, decentralized, and inde-
where B ij is a constant derived using nominal bus voltages and pendent coordination). Tractable modeling of the underlying
line reactance. Thus, since these variables can be measured lo- physical attributes of the system is imperative to efficiently
cally, there is no need to broadcast these dual variables to ac- integrate highly unpredictable power components. Various
tuating nodes, as actuation updates can be performed using
these local measurements. Although this method allows for the
establishment of optimality with respect to simplified dispatch
or load control for frequency control, it does not account for re- Table 5. A summary of independent coordination techniques.
active-power and bus-voltage magnitudes, which are important
Method Optimality Feasibility Convergence
constituents in decision making in the modern power grid.
Droop control No guarantee Feasible Not relevant
Primal-dual Guaranteed No guarantee Asymptotic
Machine learning dynamics for voltage
Machine learning is currently a prevalent area of research Machine learning No guarantee Feasible Not relevant
and is widely applied in the context of droop control for volt-

IEEE Signal Processing Magazine | May 2019 | 99


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
synchronous generators, and/or IPPs. A DN is composed of
Table 6. A summary of hierarchical framework.
buses that represent a set of residential, industrial, or com-
Tier Structure Topology Horizon Complexity mercial consumers. Bulk synchronous generators consist of
1 TN, microgrid Decentralized ~10 min O (n) active controllers that adjust mechanical input to vary real/
2 DN, IPP Distributed ~10 s O (k ) reactive-power output as necessary. IPPs can be large farms
2 Bulk generation, Centralized ~10 s O (m p) consisting of many solar panels or wind turbines with invert-
buildings ers that can be controlled to supply real/reactive power based
3 Governor, inverter Independent ~10 ms O (c )
on available generation capacities. Thus, the coordination
problem can be narrowed down from a high-level context to
grid stakeholders in the power grid enable the natural decom- a specific case.
position of the power system into manageable counterparts. Moreover, the coordination horizon is unique for various
Decision-making topologies are associated with advantages coordination applications. For instance, in the planning stage,
and trade­offs that render these suitable for a wide variety of the coordination period will be lengthier (in the scale of min-
grid applications. Thus, these must be cohesively combined utes) than primary control processes that entail fast-acting
into a grid-coordination framework that enables the seam- responses (in the scale of milliseconds). Also, the coordina-
less plug-and-play integration of power components into the tion structure will be different for each stakeholder due to the
system and adaptive response by actuating nodes to keep the inherent management attributes of these components. At the
system efficient and the grid resilient. For this, a hierarchical TN or microgrid level, due to deregulation and the participa-
approach is needed. tion of independent suppliers/consumers, decentralized coor-
The main principles for designing a hierarchical grid coor- dination is appropriate since no central authority is necessary.
dination framework are called abstraction and decoupling. In the DN, distributed coordination is appropriate due to the
The optimal power-flow problem Pc can be constructed for presence of the EPU with access to aggregate system measure-
the entire power grid where the cost of operation of individual ments. Within individual consumers (e.g., buildings and so on)
components is minimized while adhering to complex noncon- and bulk-generation systems, centralized coordination can be
vex grid constraints. If every single component of the power used to manage local appliances or mechanical systems, such
grid (e.g., power lines and consumers/suppliers) is included as governors. To adapt to transience in the system, independent
in this problem, then the problem will involve millions of coordination can be leveraged to maintain operations around
variables related to each other in a highly nonlinear manner. optimal setpoints computed in higher layers. This is summa-
Since directly solving this problem is practically impossible, rized in Table 6, where k, p, and c are constants, and n and m
the coordination problem must be divided into tractable sub- reflect the number of entities in the system.
problems that are more manageable and contained. This is In the hierarchical system defined in Figure 14, it is clear
established by leveraging the natural hierarchy of the power that the abstraction and coordination interval increases when
grid. The TN is composed of buses that represent DNs, bulk ascending the framework. Thus, in the higher tiers, planning

Short-Term Transmission/Microgrid
Total Mesh Network Short-Term
Forecast Setpoints (Decentralized) Setpoints Forecast Tier 1
Decreasing Coordination Horizon
Increasing Abstraction

Distribution/IPPS Aggregate Bulk Generation/


Radial Network Measure Setpoints Buildings Tier 2
(Distribution) (Centralized)

Governor/Inverter/
Appliance Tier 3
(Independent)

Actuation Monitoring
Multiple Components of the Same Type

FIGURE 14. The detailed hierarchical framework.

100 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
is prevalent, and, in the lower tiers, actuation is the primary The first step in the construction of this grid-coordination
focus. In the subsequent tiers, optimal setpoints computed framework by way of signal processing is to model system
for the coordination problem formulated for the underlying interactions tractably. Relaxations and approximations must
system are refined based on the decoupling and decompo- retain the discerning characteristics of physical relations
sition in effect. The coordination interval allocated accom- among power components while enabling scalability so that
modates the complexity of the coordination problem at hand. these can be used to construct signals that summarize gen-
For instance, at the TN level, decentralized coordination has eral trends in the system with respect to the physical grid.
a complexity of O (n) . If each signaling iteration takes place Realistic formulations available in the state of the art of
every 10 ms to account for the communication latency in the these relations in four specific constituents of the power grid
order of milliseconds, then 1,000 iterations can take place (TNs, DNs, microgrids, and power consumers) are presented.
within the 10-min coordination horizon. As illustrated in Fig- Then, these are leveraged in four different decision-making
ure 12, convergence in a 2,736-bus system occurs within 100 topologies that govern the construction and exchange of
iterations. Thus, the allocated coordination horizon is more monitoring/control signals that facilitate adaptive actuation
than sufficient for computing optimal setpoints for individual by individual power entities. The signals convey generalized
TN buses every 10 min. This optimal setpoint represents the information about the external environment. This informa-
real-power injection into the TN bus by either DNs, IPPs, and/ tion is used to perform adaptive actuation while heeding local
or bulk synchronous systems that reside in the subsequent infrastructure limits. Finally, these decision-making topolo-
tier. Then these systems coordinate local entities, such as gies are organized into a hierarchical framework for specific
power consumers and DGs, to maintain this optimal setpoint. power-system components that capitalize on abstraction and
These entities, such as power consumers, compute the opti- decomposition derived using a tractable optimal power-flow
mal demand setpoint over 10-s intervals. Individual elements formulation for signal processing that enables all-encom-
residing within the consumer premises locally actuate to passing grid coordination. This unified framework enables
maintain this refined setpoint via independent coordination. the seamless plug-and-play integration of heterogeneous and
As the longest coordination interval is 10 min, fluctuation in intermittent power-system components at all levels of the
the demand and generation can be accurately accommodated smart grid while accounting for the underlying physical attri-
via short-term forecasts. butes of the system.
Thus, the proposed hierarchical framework allows for the As future work, we intend to extend this hierarchical signal-
fluid coordination of individual power entities by way of abstrac- processing framework to detect and mitigate cyberphysical
tion and decomposition. This flexibility and cohesiveness in attacks, ensure small-signal/transient stability, and address
coordination can effectively accommodate the continuously power-quality issues. Moreover, it is also of interest to identify
changing landscape of the modern power grid. Although hierar- how the signal processing framework proposed in this article
chical management is not a unique concept (e.g., [50] and [51]), for the energy sector can be extended to other domains.
our work is a novel departure due to the granularity of optimiza-
tion, scalability, and the integration of various communication Authors
topologies based on the underlying physical constraints. Pirathayini Srikantha ([email protected]) received her
The hierarchical coordination framework presented in this B.A.Sc. degree in systems design engineering in 2009 and her
article focuses on the power infrastructure. The main elements M.A.Sc. degree in electrical and computer engineering in
considered in the design of this framework are the structure of 2013, both from the University of Waterloo, Canada. She
the network being coordinated, the tractable relaxation of the received her Ph.D. degree in electrical and computer engineer-
system model, and the decomposition of the coordination prob- ing from the University of Toronto, Canada, in 2017. She is an
lem into simpler subproblems. The process used in this article to assistant professor in the Department of Electrical and
identify common trends and underlying distinguishing patterns Computer Engineering at Western University, Canada. She is a
for coordination purposes can be readily applied in any interde- certified Professional Engineer in Ontario, Canada. Her
pendent processes or networked systems, such as manufactur- research interests include large-scale optimization and distrib-
ing processes, communication networks (e.g., software-defined uted control in the power grid. Her research efforts have been
systems), and water-power flow systems. Thus, hierarchical recognized in premier conference venues. She is a Member of
coordination is transferrable to a wide variety of applications. the IEEE.
Deepa Kundur ([email protected]) received her
Conclusions and future directions B.A.Sc., M.A.Sc., and Ph.D. degrees all in electrical and com-
The rapidly evolving nature of the modern power grid neces- puter engineering in 1993, 1995, and 1999, respectively, from
sitates leveraging signal processing for all-encompassing co- the University of Toronto, Canada, where she is a professor of
ordination among diverse actuating power components. Given electrical and computer engineering and serves as the chair of
the complex, connected, and collaborative nature of the cyber- the Division of Engineering Science. Her research interests
physical smart grid, a unified signal processing framework is span the fields of signal processing, cyberphysical systems, and
essential to determine and devise efficient grid dynamics and complex dynamical networks. She is an author of more than
operations while incorporating physical constraints. 200 journal and conference papers and has served as the

IEEE Signal Processing Magazine | May 2019 | 101


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.
general chair for a variety of technical meetings, including [26] P. Palensky and D. Dietrich, “Demand side management: Demand response,
intelligent energy systems, and smart loads,” IEEE Trans. Ind. Informat., vol. 7,
several GlobalSIP symposia. She is a Fellow of the IEEE and no. 3, pp. 381–388, 2011.
the Canadian Academy of Engineering. She also serves on the [27] P. Srikantha and D. Kundur, “A novel evolutionary game theoretic approach to
Advisory Board of IEEE Spectrum. real-time distributed demand response,” in Proc. IEEE Power & Engineering
Society General Meeting, 2015, pp. 1–5.
[28] M. Pham-Hung, P. Srikantha and D. Kundur, “A secure cloud architecture for
References data generated in the energy sector,” in Proc. EAI Int. Conf. Smart Grids, 2015, pp.
[1] H. Farhangi, “The path of the smart grid,” IEEE Power Energy Mag., vol. 8, no. 374–383.
1, pp. 18–28, 2010. [29] H. Quan, D. Srinivasan, and A. Khosravi, “Short-term load and wind power
[2] H. Gharavi, A. Scaglione, M. Dohler, and X. Guan, “Technical challenges of the forecasting using neural network-based prediction intervals,” IEEE Trans. Neural
smart grid: From a signal processing perspective,” IEEE Signal Process. Mag., vol. Netw. Learn. Syst., vol. 25, no. 2, pp. 303–315, 2014.
29, no. 5, pp. 12–13, 2012. [30] D. Kundur, X. Feng, S. Liu, T. Zourntos, and K. L. Butler-Purry, “Towards a
[3] J. D. Glover, T. Overbye, and M. Sarma, Power System Analysis and Design. framework for cyber attack impact analysis of the electric smart grid,” in Proc.
Boston: Cengage Learning, 2016. IEEE Smart Grid Communication Conf., 2010, pp. 1–6.
[4] T. Cutsem and C. Vournas, Voltage Stability of Electric Power Systems. New [31] P. Srikantha and D. Kundur, “Distributed optimal of dispatch in sustainable
York: Springer, 2008. generation systems via dual decomposition,” IEEE Trans. Smart Grid, vol. 6, no. 5,
[5] G. Kortuem, F. Kawsar, V. Sundramoorthy, and D. Fitton, “Smart objects as pp. 2501–2509, 2015.
building blocks for the Internet of Things,” IEEE Internet Comput., vol. 14, no. 1, [32] S. Boyd, L. Xiao, and A. Mutapcic, “Subgradient methods,” Lecture notes for
pp. 44–51, 2010. EE393o, Stanford Univ., Palo Alto, CA. 2003.
[6] A. Finamore, V. Galdi, V. Calderaro, A. Piccolo, G. Conio, and S. Grasso, [33] D. Bertsekas, A. Nedic, and A. E. Ozdaglar, Convex Analysis and
“Artificial neural network application in wind forecasting: An one-hour-ahead Optimization. Nashua, NH: Athena Scientific, 2003.
wind speed prediction,” in Proc. IET Int. Conf. Renewable Power Generation, [34] P. Srikantha and D. Kundur, “Distributed demand curtailment via water-fill-
2016, pp. 1–6. ing,” in Proc. IEEE Smart Grid Communication Conf., 2015, pp. 1–5.
[7] P. Srikantha and D. Kundur, “Hierarchical signal processing for tractable power [35] W. H. Sandholm, Population Games and Evolutionary Dynamics.
flow management in electric grid networks,” IEEE Trans. Signal Informat. Process. Cambridge, MA: MIT Press, 2009.
Netw., 2018. doi: 10.1109/TSIPN.2018.2858750.
[36] P. Srikantha and D. Kundur, “Resilient distributed real-time demand response
[8] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, UK: via population games,” IEEE Trans. Smart Grid, vol. 8, no. 6, pp. 2532–2543,
Cambridge Univ. Press, 2004. 2017.
[9] J. A. Taylor, Convex Optimization of Power Systems. Cambridge, UK: [37] P. Srikantha and D. Kundur, “A game theoretic approach to real-time robust
Cambridge Univ. Press, 2015. distributed generation dispatch,” IEEE Trans. Ind. Informat., vol. 13, no. 3, pp.
[10] K. G. Murty and S. N. Kabadi, “Some NP-complete problems in quadratic and 1006–1016, 2016.
nonlinear programming,” Math. Programming, vol. 39, no. 2, pp. 117–129, 1987. [38] P. Srikantha and D. Kundur, “Real-time integration of intermittent generation
[11] S. Diamond, R. Takapoui, and S. Boyd, “A general system for heuristic solu- with voltage rise considerations,” IEEE Trans. Sustain. Energy, vol. 8, no. 3, pp.
tion of convex problems over non-convex sets,” Optimization Methods Software, 938–952, 2016.
vol. 33, no. 1, pp. 165–193, 2018. [39] Q. Peng and S.H. Low, “Distributed algorithm for optimal power flow on an
[12] D. Lee, P. Srikantha, and D. Kundur, “Secure operating region simplification in unbalanced radial network,” in Proc. IEEE Conf. Decision and Control, 2015,
dynamic security assessment,” in Proc. IEEE Smart Grid Communication Conf., pp. 1–6.
2015, pp. 79–84. [40] C. Chen, J. Wang, and S. Kishore, “A distributed direct load control approach
[13] A. Von Meier, Electric Power Systems: A Conceptual Introduction. Hoboken, for large-scale residential demand response,” IEEE Trans. Power Syst., vol. 29,
NJ: Wiley, 2006. no. 5, pp. 2219–2228, 2014.
[14] Ontario’s Power Authority. (2017). Ontario’s long-term energy plan. Ministry of [41] B. Golub and M. O. Jackson, “Network structure and speed of learning,” Ann.
Energy, Northern Development and Mines, Toronto, ON, Canada. [Online]. Economics Statist., vol. 107, no. 108, pp. 33–48, 2012.
Available: https://www.ontario.ca/page/ontarios-long-term-energy-plan [42] S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “Distributed optimiza-
[15] R. D. Zimmerman, C. E. Murillo-Sanchez, and R. J. Thomas, “MATPOWER: tion and statistical learning via the alternating direction method of multipliers,”
Steady-state operations, planning and analysis tools for power systems research and Foundations Trends Mach. Learning, vol. 3, no. 1, pp. 1–122, 2010.
education,” IEEE Trans. Power Syst., vol. 26, no. 1, pp. 12–19, 2011. [43] D. Monderer and L. S. Shapley, “Potential games,” Games Economic
[16] Z. Yang, H. Zhong, Q. Xia, and C. Kang, “A novel network model for optimal Behaviour, vol. 14, no. 1, pp. 124–143, 1996.
power flow with reactive power and network losses,” Electric Power Syst. Res., vol. [44] T. Ui, “Discrete concavity for potential games,” Int. Game Theory Rev., vol.
144, no. 1, pp. 63–71, 2017. 10, no. 1, pp. 1–8, 2008.
[17] Z. Yang, H. Zhong, Q. Xia, A. Bose, and C. Kang, “Optimal power flow based [45] A. M. Eldurssi and R. M. O’Connell, “A fast non-dominated sorting guided
on successive linear approximation of power flow equations,” IET Generation, genetic algorithm for multi-objective power distribution system reconfiguration
Transmission Distribution, vol. 10, no. 14, pp. 3654–3662, 2016. problem,” IEEE Trans. Power Syst., vol. 30, no. 2, pp. 593–601, 2015.
[18] C. Coffrin and P. V. Hentenryck, “A linear-programming approximation of AC [46] V. Gungor, D. Sahin, T. Kocak, S. Ergut, C. Buccella, C. Cecati, G. Hancke,
power flows,” INFORMS J. Computing, vol. 26, no. 4, pp. 718–734, 2014. and D. Kundur, “Smart grid technologies: Communication technologies and stan-
[19] J. Pillai, P. Thogersen, J. Moller, and B. Bak-Jensen, “Integration of electric dards,” IEEE Trans. Ind. Informat., vol. 7, no. 4, pp. 529–539, 2011.
vehicles in low voltage Danish distribution grids,” in Proc. IEEE Power and Energy [47] C. Zhao, U. Topcu, N. Li, and S. Low, “Design and stability of load-side prima-
Society General Meeting, 2012, pp. 1–8. ry frequency control in power systems,” IEEE Trans. Autom. Control, vol. 59, no. 5,
[20] C. W. Gellings, M. Samotyj, and B. Howe, “The future’s smart delivery sys- pp. 1177–1189, 2014.
tem,” IEEE Power Energy Mag., vol. 2, no. 5, pp. 40–48, 2004. [48] H. Kakigano, Y. Miura, and T. Ise, “Distribution voltage control for DC
[21] P. Srikantha, C. Rosenberg, and S. Keshav, “An analysis of peak demand microgrids using fuzzy control and gain-scheduling technique,” IEEE Trans. Power
reductions due to elasticity of domestic appliances,” in Proc. ACM/IEEE Int. Conf. Electron., vol. 28, no. 5, pp. 2246–2258, 2013.
Future Energy Systems, 2012, pp. 1–11. [49] R. R. Negenborn, A. G. Beccuti, T. Demiray, S. Leirens, G. Damm, B. De
[22] S. H. Low, “Convex relaxation of optimal power flow part I: Formulations and Schutter, and M. Morari, “Supervisory hybrid model predictive control for volt-
equivalence,” IEEE Trans. Control Netw. Syst., vol. 1, no. 1, pp. 15–27, 2014. age stability of power networks,” in Proc. American Control Conf., 2007,
[23] L. Gan, U. Topcu, and S. Low, “Exact convex relaxation of optimal power flow pp. 5444–5449.
in radial networks,” IEEE Trans. Autom. Control, vol. 60, no. 1, pp. 72–87, 2015. [50] R. J. Sanchez-Garcia, M. Fennelly, S. Norris, N. Wright, G. Niblo, J. Brodzki,
[24] D. E. Olivares, A. Mehrizi-Sani, A. H. Etemadi, C. A. Canizares, R. Iravani, and J. W. Bialek, “Hierarchical spectral clustering of power grids,” IEEE Trans.
M. Kazerani, A. H. Hajimiragha, O. Gomis-Bellmunt, M. Saeedifard, R. Palma- Power Syst., vol. 29, no. 5, pp. 2229–2237, 2014.
Behnke, G. A. Jimenez-Estevez, and N. D. Hatziargyriou, “Trends in microgrid [51] P. Tian, X. Xiao, K. Wang, and R. Ding, “A hierarchical energy management
control,” IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 1905–1919, 2014. system based on hierarchical optimization for microgrid community economic oper-
[25] A. Etemadi, E. J. Davison, and R. Iravani, “A decentralized robust control ation,” IEEE Trans. Smart Grid, vol. 7, no. 5, pp. 2230–2241, 2016.
strategy for multi-DER microgrids–Part I: Fundamental concepts,” IEEE Trans.
Power Del., vol. 27, no. 4, pp. 1843–1853, 2012.  SP

102 IEEE Signal Processing Magazine | May 2019 |


Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on February 19,2022 at 11:19:34 UTC from IEEE Xplore. Restrictions apply.

You might also like